10.07.2015 Views

Web Mining and Social Networking: Techniques and ... - tud.ttu.ee

Web Mining and Social Networking: Techniques and ... - tud.ttu.ee

Web Mining and Social Networking: Techniques and ... - tud.ttu.ee

SHOW MORE
SHOW LESS
  • No tags were found...

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

8.1 User-based <strong>and</strong> Item-based Collaborative Filtering Recommender Systems 173P u,i = ∑ ( )j∈I i sim(i, j) × Ru, j∑ j∈Ii (|sim(i, j)|)(8.4)where I i is the set items that are similar to item i, || denotes the size of the set. The illustrationof item-based collaborative filtering algorithm is depicted in Fig.8.2. Actually it is furtherconcluded that the weighted sum approach predicts the overall rating preference of the targetitem for the user by referring to the similar items determined by analyzing the co-rating data.Fig. 8.2. The item-based collaborating filtering recommendation process [218]RegressionThis approach follows the similar prediction principle but with different calculation strategy.Instead of using the ”raw” rating scores of similar items in computing the weighted sum,this approach uses an approximation of the ratings based on regression model. The initiatedmotivation behind is the fact that in practice even the calculated similarity betw<strong>ee</strong>n two itemsusing cosine or correlation is quite high, the rating vectors may be distant (in Euclidean space).In this case using the raw rating scores of the similar items may result in the poor predictionperformance. An intuitive solution is to use formulated rating scores rather than the raw onesof similar item for computing the weighted sum. This process is realized by a linear regressionmodel. If the target item <strong>and</strong> the similar items are denoted by R i <strong>and</strong> R j respectively, the linearregression model can be expressed asR ′ j = αR i + β + ε (8.5)The regression model parameters α <strong>and</strong> β are empirically determined. ε is the regression error.In comparison to user-based collaborative filtering algorithms, one significant advantageof model-based algorithms is the scalability of the recommendations. In user-based CF (orneighborhood-based CF) systems, the neighborhood computation is very time consuming especiallyin case of larger e-commerce sites, making it almost unsuitable in real time applications.However, model-based CF systems have the potential to be employed recommendersystems operating at a large scale. The solid solution to this problem is to separate the wholerecommendation process into two stages: one offline stage <strong>and</strong> one online stage. The model

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!